Daniel Kahneman · Thinking, Fast and Slow
A long book on one page. Read the gold box if you have one minute. Read the red and green pairs if you have five. The blue boxes are the experiments that made each idea stick.
What it is, and whether it is for you.
A summary of roughly forty years of research into how people actually judge and decide, written by the psychologist who, with Amos Tversky, did most of it. The central device is two characters: System 1, fast, automatic and always running, and System 2, slow, effortful and easily tired. Kahneman is explicit that these are useful fictions rather than parts of the brain.
The book builds in five parts: the two systems, the heuristics that produce predictable errors, the overconfidence that follows from them, how we actually choose under risk, and finally the gap between the self that lives an experience and the self that remembers it. It is written for the intelligent general reader and assumes no statistics.
If you remember nothing else, remember these.
Each part sets up the next: the systems produce the heuristics, the heuristics produce the overconfidence, and so on.
Introduces System 1 and System 2, attention and effort, and the associative machinery that produces impressions before you have decided anything. Covers cognitive ease, the way the mind jumps to conclusions, and substitution.
System 2 believes it is doing the work. Usually it is signing off on System 1's output.
The shortcuts that stand in for real statistical reasoning: anchoring, availability, representativeness, insensitivity to sample size and to base rates, and the failure to expect regression to the mean.
The errors are not random. They are systematic, which means they can be anticipated.
Why we overestimate how well we understand the past and how well we can predict the future. Covers hindsight bias, the illusion of validity, when expert intuition can be trusted, and the planning fallacy.
Expert intuition is real, but only where the environment is regular enough to learn from and feedback is quick and clear.
Prospect theory, reference points, loss aversion, the endowment effect, the overweighting of rare events, mental accounting and sunk cost, and how framing changes decisions without changing the facts.
People do not evaluate outcomes. They evaluate changes from a reference point, and losses hurt more than gains please.
The split between the self that lives through an experience moment by moment and the self that files a report afterwards. Covers the peak-end rule, duration neglect and the focusing illusion.
You do not choose between experiences. You choose between memories of experiences.
Red shows how the mind goes wrong. Green shows the correction that actually works.
What it looks like
The first number mentioned drags every later estimate toward it, including numbers you know to be arbitrary and numbers you consciously reject.Your move
Generate your own estimate before you hear anyone else's, and write it down. If an anchor has already landed, deliberately argue the opposite case before settling.Why it works: The pull is automatic, so it cannot be resisted by intention. It can only be pre-empted or counterweighted.
Before we look at their number, what is ours?
What it looks like
You judge how common something is by how easily examples come to mind. Recent, vivid and emotionally charged events feel far more frequent than they are.Your move
Ask for the base rate before discussing the recent case. Count, or find someone who has counted.Why it works: Ease of recall is driven by drama and recency, neither of which correlates with frequency.
How often has this actually happened, across how many cases?
What it looks like
A hard question gets silently replaced by an easier one, and the answer to the easy question is delivered with the confidence appropriate to the hard one.Your move
State the question you are actually answering, out loud. Compare it with the question you were asked.Why it works: Substitution survives on being invisible. Naming it is most of the fix.
What question am I actually answering here?
What it looks like
A confident conclusion built entirely on the evidence that happened to be available, with no weight given to what is missing.Your move
Before deciding, list what you would need to know to be confident, then mark which of those you actually have.Why it works: It converts absent evidence from invisible into visible, which is the only way System 1 can register it.
What would we need to know that we do not know?
What it looks like
Plans built from the inside, imagining this project going well, and consistently landing over time and over budget.Your move
Take the outside view. Find the reference class of similar past projects and start from their actual distribution of outcomes.Why it works: Your case feels unique from inside. Statistically it rarely is, and the base rate already contains the surprises you have not imagined.
What happened to the last ten projects like this one?
What it looks like
Once you know the outcome, the path to it looks obvious, and you cannot reconstruct how uncertain you genuinely were beforehand.Your move
Record predictions and their reasoning in writing, before the outcome. Score them later against what you actually wrote.Why it works: Memory rewrites prior beliefs to match known outcomes. Only a contemporaneous written record survives the rewrite.
What did we write down at the time?
What it looks like
High confidence in a judgement produced by a coherent impression, in a setting with no reliable feedback, such as an unstructured interview.Your move
Break the judgement into a few independent, defined factors, score each separately, and only then form an overall view.Why it works: Decomposing prevents a single strong impression from colouring every dimension at once.
Let us score each factor separately before we discuss the person.
What it looks like
Each decision judged on its own, one at a time, which makes sensible individual risks look unacceptable.Your move
Set a policy that covers the whole class of decisions, and apply it to every instance rather than re-deciding each time.Why it works: Broad framing lets the arithmetic of many small independent bets work in your favour, which single-instance thinking destroys.
What is our rule for all decisions of this type?
The studies the arguments rest on. These are from the published research, not invented for this page.
Tversky and Kahneman · the conjunction fallacy
Participants read a description of Linda: single, outspoken, deeply concerned with discrimination and social justice. They were then asked which was more probable, that Linda is a bank teller, or that she is a bank teller who is active in the feminist movement.
The turn: A large majority chose the second, which cannot be more probable than the first, because it is a subset of it. The detailed story was more representative of the description, and representativeness beat logic.
Lesson: Adding plausible detail makes a scenario feel more likely while making it mathematically less likely.
Anchoring
Participants watched a wheel spin and stop on a number that was, unknown to them, rigged to land on 10 or 65. They were then asked what percentage of United Nations member states were African.
The turn: Those who saw 10 guessed around 25 per cent. Those who saw 65 guessed around 45 per cent. A number they had just watched come from a spinning wheel moved their estimate substantially.
Lesson: Anchors work even when the anchor is visibly irrelevant and the person knows it.
Regression to the mean
Israeli flight instructors told Kahneman that praising a cadet for a good manoeuvre was followed by worse performance, while criticising a bad one was followed by improvement. They concluded that praise hurts and criticism helps.
The turn: Both patterns are what regression to the mean produces on its own. An exceptional performance is usually followed by a more average one regardless of what anyone says.
Lesson: Wherever performance is noisy, the world will appear to reward punishment and punish reward.
Peak-end rule and duration neglect
Participants held a hand in painfully cold water for 60 seconds. In a second trial they did the same 60 seconds, then kept the hand in for a further 30 seconds while the water was warmed very slightly, still unpleasant but less so.
The turn: Asked which trial they would repeat, most chose the longer one, which contained strictly more total pain. The gentler ending dominated the memory, and the extra duration barely registered.
Lesson: Memory records peaks and endings, not totals, and memory is what you use to choose next time.
Techniques you can use straight away.
Before committing, have the group imagine it is a year later and the decision failed badly. Each person writes a short history of that failure, independently, before anyone speaks.
Use when: Any significant decision where the group is converging quickly and agreement feels comfortable.
Identify the class of comparable past cases, find their actual distribution of cost, duration or outcome, start from that, and adjust only for differences you can specify.
Use when: Any estimate of how long something will take or how much it will cost.
Choose a handful of factors in advance, define what each means, score them one at a time and independently, and compute the total before forming an overall impression.
Use when: Hiring, vendor selection, grant assessment, anywhere a global impression would otherwise dominate.
Before the outcome is known, record the decision, the reasoning, the alternatives rejected and the confidence level. Review it after the result.
Use when: Recurring decisions where you want genuine feedback rather than a memory reshaped by hindsight.
Decide the rule for a whole category of decisions once, in the calm, then apply it to each case rather than re-opening the question under pressure.
Use when: Repeated small risks, where narrow framing and loss aversion would otherwise make you too cautious.
Ask what baseline the parties are comparing against. Reframe the same facts as a gain from one reference point and a loss from another, and see whether preferences flip.
Use when: Negotiations, pricing, and any proposal meeting resistance that the numbers do not explain.
Slow thinking is expensive, so spend it where it changes the answer.
One small action a day.
Write down three predictions you are currently making at work, each with a confidence level and a date to check.
In your next meeting, before any numbers are shared, write your own estimate on paper first.
Take one active project and find three comparable past projects. Compare their actual duration with your current plan.
Run a ten-minute premortem on a decision that currently feels settled.
Take one recurring judgement you make by impression and break it into three defined factors scored separately.
Find one belief you hold that rests on a vivid recent example, and go looking for the base rate.
Re-read Day 1. Note which predictions you would now revise, and what made you revise them.
Tap the ones that sound like your last month.
0 of 7 selected.
Five minutes with these is worth more than a second read.
★★★★★
The foundational text on judgement, written by the person who did the research, and honest about what it does not settle.
Bottom line: You cannot think your way out of fast thinking, and the book does not pretend otherwise. What you can change is the process around the decision: estimate before you are anchored, start from the record of comparable cases rather than from your picture of this one, assume failure and write its history before you commit, and keep a dated note of what you believed and why. The value is not in spotting biases in yourself in the moment, which almost nobody manages. It is in building the few procedures that catch them on your behalf.